The four HaiPick systems, compared
Hai Robotics sells four base systems, not three: HaiPick System 1, System 2, System 3 and HaiPick Climb. A project starts by choosing one of them and then tailoring it:
We start by identifying a base system that fits your requirements. From there, modular elements are added to tailor the solution for peak efficiency and performance. — system components catalog
Every figure below is quoted from hairobotics.com and machine-checked. Where a figure comes from a case study it names the customer. Anything that is my own reasoning is marked as such — it is not Hai's claim, and should not be presented to them as one.
Each system links to its entry in the product base, which carries the quote behind every number.
The short answer
| If the binding constraint is… | Take | Because |
|---|---|---|
| A straightforward tote operation, standard racking, moderate volume | System 1 | One robot type, 2 totes deep, up to 12 m, any container or none. The least to commission and the least to go wrong |
| Bulky or palletised goods stored with the totes | System 2 | Adds heavy-duty companion AMRs for bulk and pallets, so mixed inventory stays in one system |
| Maximum density and high throughput on a large site | System 3 | Triple-deep Chain-Pick storage, 43,000 totes per 1,000 m², and fast AMRs doing the horizontal legs at 4 m/s |
| Density with a simple build and quick go-live, totes only | HaiPick Climb | Robots climb the racking itself: 45,000 totes per 1,000 m², a tote at a workstation in under 2 minutes |
What each one is
HaiPick System 1 — the core goods-to-person system. ACRs drive the floor, climb into standard racking "up to 12 meters (39+ feet) high and 2 totes deep", and "batch deliver up to 9 containers simultaneously to a human-operated picking station". It is container-agnostic to the point of needing none at all: "In System 1, products can even be stored without containers." Claims 99%+ picking accuracy and labour cost down "by up to 67%".
HaiPick System 2 — System 1 plus bulk. It "achieves all the benefits of System 1 — with added bulk, palletized, and mixed-type storage capabilities", by adding "Heavy-Duty Companion AMRs for bulk and palletized goods picking" (the K600 and K1000, 600 kg and 1000 kg). The catch is structural: the racking must have "enough clearance at the floor level to maneuver bulk items".
HaiPick System 3 — the density and throughput system, and the one with two robot classes. ACRs do the vertical work ("picking and storing totes up to 12 meters high, in triple-deep configurations, and handling loads up to 50 kg") while companion AMRs do the horizontal work "at speeds of up to 4 m/s". Density comes from Chain-Pick, which removes the gap between totes front to back: "0 mm front-to-back gaps", "40 mm* side-to-side gaps", "storing up to 43,000 totes in 1,000 square meters". It also handles the widest range of containers — "totes, cartons, and trays simultaneously in one warehouse", 300×300 mm to 850×650 mm — and supports trays so cartons keep their original packaging. HaiVest lets staff walk in while it runs.
HaiPick Climb — a different machine entirely. Instead of a floor robot that reaches up, a HaiClimber "moves freely up, down, and beneath racks at up to 4 meters per second", climbing at 1 m/s. Double-deep, "up to 45,000 totes within a 1,000 m²", a tote delivered "in under 2 minutes", and a published ceiling of "Fleet sizes up to 3,000 robots". Hai positions it as the simple one: "Simple implementation and quick go-live with minimal floor requirements" and "fast 30-minute training".
Published figures, side by side
"—" means the site does not publish it, which is itself worth knowing. The live version of this table is at /products/compare.
| System 1 | System 2 | System 3 | Climb | |
|---|---|---|---|---|
| Robots | ACRs | ACRs + heavy-duty AMRs | ACRs + fast-transit AMRs | HaiClimbers |
| Max height | 12 m | 12 m | 12 m | 12 m |
| Storage depth | 2 totes deep | 2 totes deep | up to triple-deep, can vary by aisle | double-deep |
| Density | — | — | 43,000 totes / 1,000 m² | 45,000 totes / 1,000 m² |
| Throughput | — | — | — | 4,000 totes/h per 1,000 m² |
| Payload per container | — | — | 50 kg | — |
| Container range | any container, or none | any container, or none | 300×300–850×650 mm; totes, cartons, trays | totes, trays, cartons in original packaging |
| Cases per trip | 9 | — | — | — |
| Picking accuracy | 99%+ | 99%+ | 99%+ | — |
| Labour reduction | up to 67% | — | 67% | — |
| Delivery time | — | — | — | under 2 minutes |
| Certification | CE, NRTL | CE, NRTL | — | — |
Two things the table hides:
- The 12 m is the same everywhere, so height is rarely the deciding factor between them. Depth, container mix and robot type are.
- Density is only published for System 3 and Climb. The 43,000 and 45,000 figures are close enough to be a wash; the difference between them is how the density is reached — triple-deep with Chain-Pick versus double-deep with narrow aisles.
What the deployments actually look like
The 89 case studies name a system 32 times: System 1 in 10, System 2 in 1, System 3 in 16, Climb in 5. That distribution is itself a signal — System 3 is where the recent, large projects are, and System 2 has almost no public track record.
| System | Sites | Storage locations | Outbound throughput | Racking height | Industries |
|---|---|---|---|---|---|
| System 1 | 10 | 3,820 → 40,000 | 78 → 1,170 totes/h | 4.25 → 12 m | electronics, healthcare, retail, automotive, 3PL, MRO |
| System 2 | 1 | 4,500 tote + 400 pallet | — | 5.7 m | electronics manufacturing |
| System 3 | 16 | 8,000 → 625,000 | 110 → 9,400 totes/h | 3.2 → 12 m | apparel, healthcare, 3PL, retail, automotive, beauty |
| Climb | 5 | 9,000 → 120,000 | 230 → 2,400 totes/h | 5.7 → 10 m | apparel, e-commerce, electronics |
Representative deployments:
| System | Customer | Fleet | Locations | Outbound |
|---|---|---|---|---|
| 1 | CDW UK | 11 ACRs (12 m) | 16,000 | 300 totes/h |
| 1 | CEVA Logistics | 35 robots, 3 HaiPorts | 24,000+ | 942 totes/h |
| 1 | St. Luke's (healthcare, US) | 28 ACRs | 18,600+ | 650+ totes/h |
| 2 | GIGABYTE | 5 ACRs + 4 heavy-duty AMRs | 4,500 totes + 400 pallets | — |
| 3 | LPP Logistics | 278 ACRs + 670 AMRs | 625,000+ | 9,400+ totes/h |
| 3 | Arvato (beauty) | 72 A42TD-L-E1 + 151 K50H | 88,400+ | 2,900+ totes/h |
| 3 | RD Saúde (pharma, Brazil) | 27 high-bay ACRs of an 85-unit fleet | 62,000+ | 1,140 totes/h |
| 3 | Transmec (3PL, Italy) | 5 ACRs + 10 K50 | 13,000 | 250 totes/h |
| Climb | Anta | 125 HaiClimbers, 35 chargers | 120,000+ | 2,400 totes/h |
| Climb | Mettler Toledo | 9 HaiClimbers | 9,000+ | 230+ totes/h |
Throughput per robot — my arithmetic, not Hai's
Dividing each case's published outbound rate by its published robot count gives a first-pass sizing rule. Hai does not publish these ratios; I calculated them, and they are rough — outbound rate is a whole-system claim, and workstation count, travel distance and order profile all move it.
| System | Sites used | Totes/h per ACR or climber |
|---|---|---|
| System 1 | CEVA 27, Inteq 27, CDW 27, KEC 26, St. Luke's 23 | 23–27, remarkably consistent |
| System 3 | Transmec 50, RD Saúde 42, Arvato 40, LENOVO 40, LPP 34, KOHLER 28 | 28–50 |
| Climb | Leading apparel 35, Mettler 26, e-commerce 22, Anta 19, LI-NING 16 | 16–35 |
What I read into this (reasoning, not fact): System 1's tight 23–27 band is what you would expect when one robot does both the vertical pick and the horizontal carry — the cycle is dominated by travel, and travel is similar everywhere. System 3 runs higher per ACR precisely because the AMRs take the horizontal leg away, letting the expensive climbing robot stay in its aisle. That is the real argument for the two-robot design, and it is worth checking against a number Hai gives you. Climb's spread is the widest, which fits a system where throughput depends on how many workstations the aisles feed.
Choosing a system
| Scenario | Likely fit | Reasoning (mine) |
|---|---|---|
| Existing building, 5–8 m clear, standard totes, moderate lines/day | System 1 | Nothing in the mix needs depth beyond 2 totes or a second robot class. Fewest interfaces to commission |
| Low clear height (under ~5 m) | System 1 or Climb | Both are deployed low — LENOVO ran System 3 at 3.2 m and KEC System 1 at 4.25 m — but paying for density you cannot build upward is wasteful |
| Cartons and bulky or palletised goods in the same inventory | System 2 | The only system that puts bulk in the same automated flow. Ask for references — there is one public case |
| Original packaging, full-case in and out, no decanting | System 3 | Tray handling and auto tray/de-tray are System 3 features; New Wave Textiles put about 76% of inbound cartons on trays |
| Very large site, high SKU count, peak-driven | System 3 | The only system with published deployments above 100,000 locations: LPP 625,000, NEPA 130,000, New Wave 127,000 |
| Space-constrained site where density is the whole problem | System 3 or Climb | 43,000 versus 45,000 totes per 1,000 m². Decide on container mix and build effort, not density |
| Fast go-live, small team, tote-only | Climb | "quick go-live with minimal floor requirements", 30-minute training; the apparel site went live in 4 weeks |
| Cold chain or regulated storage | System 1 | Zuellig Pharma runs ambient and cold zones with dedicated ACRs per zone — a zoning pattern, not a product feature |
| Layout likely to change; site may be relocated | System 1 or 3 | Inference: ACRs are floor robots working with "almost any industry-standard racking", while HaiClimber robots run on the rack structure itself. Reconfiguring a climbing system should be the bigger job. Hai does not say this — worth asking |
What is not published, and what to ask
- HaiClimber per-robot specs. No payload, no cases per trip, no tote envelope. You cannot size a Climb system from public material the way you can System 1.
- Climb density contradicts itself: 45,000 totes per 1,000 m² on the Climb and systems pages, "up to 30,000" on the racking page. Ask which is current, and at what rack height.
- System 2 has one public case. If bulk handling is the reason to buy, ask for references and for the real split of AMR and ACR work.
- Newer robot models are not on the site. Recent System 1 and System 3 cases name A42TD-L-E1, A42TD-G-E1, A42TD-G and K50H models, none of which have a product page. Ask what the current model line is — the robot pages look a generation behind the case studies.
- No throughput model. Nothing published says how many robots a target rate needs. The ratios above are my substitute; ask them for theirs.
Sources
Every quoted figure is verbatim from hairobotics.com. The system figures are checked
by poetry run python scraper/verify_products.py through the product base; the 32
case-study quotes used here were checked the same way against data/pages.jsonl.
The overview page is
About HaiPick Systems;
each system links to its own page from its product base entry.
Case figures ignore the site's boilerplate "Products" footer — see DESIGN.md for why that matters.